Mingjing Tong presented results from her parallel runs and the assimilation of tcvital wind speed data. Mingjing introduced her subject by presenting a chart listing the differences between the H050 (new HWRF baseline)/H051 (H050+initialization changes) models and her control, T050. First, H050(H051) use GFS analysis for the storm environment while T050 uses GFS forecast. The second and biggest difference between the two involves HWRF initialization, and H050 uses relocation plus size correction plus intensity correction while T050 uses full HWRF initialization only when radial velocity (Vr) data are not available and relocation when Vr are available (without size and intensity correction). Third, H050 uses an old version of GSI with an isotropic background error covariance while T050 uses the Q1FY10 GSI with anisotropic background error covariance. Finally, H050 uses satellite data while T050 uses satellite data with the addition of conventional data, pseudo-MSLP data, and airborne radar Vr (if available).

Next, Mingjing presented track (left plots) and intensity (right plots) errors for H050 (in red), H051 (in blue), and T050 (in pink) for parallel runs for Bill (03L) from the 2009 hurricane season. For track, T050 showed a degradation compared to H050 and H051, especially after 48h. However, for intensity, T050 showed the lowest errors from 24-96h compared to H050 and H051. Mingjing then presented track and intensity plots for the Bill parallel run cycles without airborne radar Vr data. For track, T050 error values are reduced compared to the previous plot and similar to H050 and H051. Intensity error for T050 is similar to the previous plot with lowest error values for all experiments from T050 at 24-72h. Track and intensity error plots for cycles of Bill with airborne Vr data were then shown. This plot includes T051 (in pink with dashed line), which includes relocation plus size correction plus radar data assimilation, and T052 (in green), which includes the full HWRF initialization plus radar data assimilation. The track plot shows T050, T051, and T052 with similar error values and all higher than errors for H050 and H051. For intensity, T050, T051, and T052 have the largest error values initially, but these values drop after 24h and become comparable to H050 and H051. Mingjing explained that the higher error values for her T050(1,2) experiments didn't seem reasonable compared to her previous results using the H209 version of HWRF. In those track and intensity plots, Mingjing's experiments were ADR1(pink) and ADR6(blue) compared to H209 (red). Both ADR1 and ADR6 had lower track and intensity errors compared to H209. While the intensity errors seen for the ADR1(6) and T050 experiments are similar, the track errors for ADR1(6) show that using radar data didn't increase the error values unlike with T050(1,2). Mingjing then showed plots of MSLP (top row) and 10m wind maximum (bottom row) for three Bill cycles with radar data. In these plots, H051 (red) and T050 (blue) are compared to best track (in black). For all plots, the low MSLP values for both T050 and H051 don't seem very consistent with the 10m max wind values. The MSLP values deepen very quickly within the first 12h while the winds don't increase nearly as fast. Mingjing speculated that this was possibly due to the wind-pressure relationship used in the HWRF.

Then, Mingjing presented parallel run results for Gustav (07L). The track error plot shows T050 values comparable to those from H050 and H051, while T050 intensity error values are lower than H050 and H051 from 24-48h, and lower than H050 for all hours except 120h. For the Gustav runs without airborne radar Vr data, results are similar to those shown in the previous plots with T050 track error similar to those for H050 and H051 and lower T050 intensity error compared to H050 except at 120h. For the Gustav runs with airborne Vr data, T051 had the highest track errors, but T052 has the lowest track errors especially from 36-96h. T051 also had the lowest intensity errors compared to T050 and T051, but it wasn't consistently better than H050 and H051. For the Gustav plots of MSLP and max 10m wind comparing H051 and T050 data, Mingjing again noted the strong pressure values for Gustav not being consistent with the max wind values.For the Ike (09L) parallel runs, T050 showed consistently lower track error values than H050 and H051. For intensity error, T050 had lower values from 12-60h, and was highest compared to H050 and H051 at 96h. In track and intensity error plots for Ike cycles without airborne Vr data, T050 showed similar results to the previous plots.

Mingjing then presented an overview of the assimilation of tcvital 17m/s wind speed data. For her work, the hurricane storm size is defined as the average radius of the outer closed isobar (ROCI--which was not used here) or of gale force (17m/s) wind. From her previous results, Mingjing found that for correcting storm size, assimilating airborne radar Vr data is not very effective. Also, multi-pass analysis wasn't tried with a different thinning grid for Vr data or with different background error covariance lengths. The tcvital data include the average radius of 17m/s wind speeds for each of the four storm quadrants. Mingjing noted that the goal of this assimilation work was to control the storm size by assimilating the 17m/s wind speed data with the tcvital data being treated as if located in the NE, NW, SE, and SW directions with the corresponding radius. In the plot showing the HWRF initialization size correction (indicated by the color contours), the color contours indicate a decrease in storm size at all levels. For the assimilation of tcvital wind speed data plot, the color contours again show the storm size reduced but mainly at the lower and middle levels. Next Mingjing showed surface pressure, 850 mb steamline and isotach, and E-W cross-section plots for the 2008083000 cycle of Gustav. Looking at the storm size from the first guess (top row) to the assimilation of both tcvital MSLP and wind speed (bottom row), the storm size reduction is clearly visible.

Mingjing then detailed her new method for combining different types of data. She first noted that little impact was seen if tcvital wind speed data were assimilated using the same error covariance length scale as that used for Vr data. Since a hurricane is a multi-scale system, different error covariance lengths may need to be used to resolve scales represented by different data. To address this issue, Mingjing generated a two-pass GSI analysis. In the first pass, tcvital wind speed and MSLP data are assimilated using larger error covariance lengths to correct storm size and intensity. In the second pass, radar data are assimilated using smaller error covariance lengths to add more detailed structures. Mingjing then presented some preliminary results from this work for Bill, Ike and Gustav. In these plots, H051 is in green, T050 is in blue, and T053 (which uses the two-pass analysis) is in red. The storm tracks for T050 and T053 are very similar and not closer to the best track than H051 for all storms. For the MSLP intensity, T050 and T053 are again very similar and slightly better than H051 for Bill (early on) and Gustav (36-72h). The same is true for maximum wind intensity plots. Mingjing indicated that the two-pass analysis did not produce much of an impact, however, there was no negative impact noticed.

To conclude, Mingjing presented her future work plans. They included further diagnosis of the results for cycles using radar data assimilation and verification of analysis using independent Vr observations. Also, further testing or tuning the different methods for combining different types of data should be completed and cycles should be run. Mingjing also planned to conduct parallel runs for three more storms: Fay (06L), Kyle (11L) and Paloma (17L) as well as parallel runs using Young Kwon's new surface physics (Cd, Ch). Finally, she planned to merge the GSI code to the subversion trunk version and start work on a hybrid system.